data.table(vector, matrix) can return duplicate column names

[MichaelChirico]

data.table(1:5, matrix(6:15, nrow = 5L))
#    V1 V1 V2
# 1:  1  6 11
# 2:  2  7 12
# 3:  3  8 13
# 4:  4  9 14
# 5:  5 10 15

Not ideal to be creating data.tables with duplicate names from the get-go... those names should be V2_1 and V2_2 i guess.

Fixes with check.names but I'm not sure creating duplicates like this by default is the right behavior; my only hesitation is I'm sure there are other more complicated cases that are still creating duplicates even if we fix this.

The fix to this specific issue can be very simple, however, just add 7 characters to this line:

https://github.com/Rdatatable/data.table/blob/master/R/data.table.R#L98

if (any(tt)) namesi[tt] = paste0("V", i, '_', which(tt))

[Atrebas]

Sowewhat related, I have noticed several cases when colnames are duplicated.
These may be some specific cases, when working with toy datasets and colnames V1, V2, ...
Some examples below for information.

library(data.table)

## case 1: setDT duplicates V1 and V2
lst = list(V1=1, V2=1, 1, 1)
setDT(lst)[] #check.names=TRUE solves this


DT = data.table(V1 = c(1L,2L),
                V2 = 1:9,
                V3 = round(rnorm(3),2),
                V4 = LETTERS[1:3])

## case 2: computation in j duplicates V1
DT[, sum(V3), by = "V1"] 

## case 3 (linked to case 2): rollup duplicates the V1 colname and bugs
rollup(DT, j = sum(V3), by = c("V4", "V2"))   # ok
rollup(DT, j = sum(V3), by = c("V4", "V1"))   # error

# Error in groupingsets.data.table(x, by = by, sets = sets, .SDcols = .SDcols,  : 
# There exists duplicated column names in the results, 
# ensure the column passed/evaluated in `j` and those in `by` are not overlapping.

# So specifying a different name works:
rollup(DT, j = .(SumV3 = sum(V3)), by = c("V4", "V1"))   # ok

[jangorecki]

@Atrebas thanks for investigation. Groupings sets functions just wraps around ordinary grouping, thus are affected.

[jangorecki]

Related to #4124

[jangorecki]

@Atrebas

• case 1: setDT is designed for speed, so the less the checks the better, any checks on quality of data/metadata should be on user side
• case 2: has a dedicated issue duplicate names on simple operation #3644
• case 3: inherits from case 2, as explained before.

[jangorecki]

This issue is caused by a fact that data.table has check.names=FALSE by default, while data.frame has check.names=TRUE. I filled #4362

[MichaelChirico]

I think this is fixed by #4363, though the breaking change means it's an "eventually" thing.